A QCW sword that comes out straight in one room can bend or loop in another, on the same machine. The cause is not known. The straightness depends on the room more than on the coil, so two builders with the same design get different arcs, and one of them goes looking for a fault that is not there.
None of the observations here are ours, because our own machine has never been switched on. They come from other builders, and the explanations attached to them are untested hypotheses.
Two defects, constantly confused
Branching is the arc growing in all directions at once. Wander is the arc curving as a whole, sometimes right round into a loop.
The two respond to frequency in opposite directions. Raising the frequency makes branching better, because the channel runs hotter, and makes wander worse. A ramped VTTC at 2 MHz makes loops tens of centimetres across.
davekni describes a third failure. He links the two ways of making an arc longer to two different kinds of branching, at opposite ends of the arc:
If length is added by increasing growth rate, branching is likely near tip. If length is added by increasing ramp time, electrostatic repulsion of the arc with itself near the base makes the path unstable, leading to branching near the base. This instability is obvious in my low frequency (100kHz) QCW experiments, but still visible in many QCW arcs towards end of ramp time (HVF 3140).
The last sentence matters. He is not describing an oddity of one 100 kHz experiment. He says self-repulsion shows up late in the ramp on ordinary machines, and late in the ramp is where a QCW gets most of its length.
Frequency does trade one thing against another, but the trade is length against stability. The high end has a figure on it: asked about swords above 500 kHz, davekni says staccato VTTCs make them at 1 to 2 MHz and that they are "Shorter arcs than most silicon or SiC ramped or QCW coils". At high frequency you get a straight arc but a shorter one. At low frequency you get length but also self-repulsion. The middle of the range is where you can have reasonable amounts of both. That explains the 200 to 500 kHz band better than the branching argument below does, and it rests on two sourced observations.
We put the long straight sword at 200 to 500 kHz, but that band comes from our own description of what a QCW arc looks like, not from a measured optimum between branching and wander. The wander side of it holds up, with the 2 MHz VTTC as evidence. The branching side does not: the list further down gives the causes of branching and frequency is not one of them, and the same table of ours calls 1 to 4 MHz a thin straight warm flame. Take the band as a description of the arc's shape, not as a setting found by trading two defects against each other.
Zigzags are a separate thing again, and the mechanism reported for them is not branching. davekni's 100 kHz QCW went from a straight path to a zigzagged one, and he reports the discussion settling on the arc repelling itself: adjacent sections of one channel pushing apart, with higher frequency reducing it (HVF 3140). This is the same space charge that sets the ceiling on ramp time. It predicts a zigzag with no branching at all, so a crooked arc and a branchy arc are not necessarily the same defect.
What people have seen
CJ has the most detailed set of observations, on HVF 2434. He moved the machine to a different spot and the arc bent. He put it near a wooden cabinet, a wardrobe in his own words, and it straightened. He lists this under "the arrangement of surrounding conductive materials such as wardrobes, tables, beds, and walls", and then says straight after that the wardrobe "is made of wood and does not conduct electricity". The contradiction is his, and it tells you something: whatever the wardrobe did, he did not think it did it by conducting.
davekni replies in the same thread, and it is the only physics anyone offers for a cupboard: "though wood is not very conductive at low frequency, it can be significantly conductive at 12MHz. And dielectric materials should have some effect even if not conductive at all". The frequency he names is 12 MHz, two orders of magnitude above where these coils run, so the conductive half of that counts for less here than it seems. The dielectric half needs no conductivity and is the part that still applies.
CJ's other observations: in a cold winter the arc bent down towards the ground; he lit the stove, the room warmed up, and the arc straightened. Near an iron frame it was straighter, though he says that was a small coil of his, not the machine the rest of the list came from (reply 9). The thickness of the grading ring hardly mattered (reply 8).
The simplest observation in any of the threads is davekni's, and it ties curvature to length instead of to the room: "even the curved arcs are comparatively straight if aborted earlier" (HVF 2397). Same coil, same room, same air; stop the ramp sooner and the arc is straighter. Whatever bends the arc needs length to do it.
Jan, after maybe fifteen years of building coils, ran into this for the first time: arcs looping downward with a diameter under a metre, sometimes in an S. Dry air went with straighter arcs, though he calls the correlation weak and says plainly that he does not really know. Temperature alone did almost nothing. Opening the garage door to about ten degrees of outside air reduced the looping (HVF 1073).
One more observation gets passed around: arcs curling almost to the floor at low atmospheric pressure, with temperature and humidity unchanged. We have not found where it was first reported. The closest thing we found is CJ including atmospheric pressure in his list of conditions his arc responded to, in the post above, and that is not the same observation. We mention it without a name because it is the only pressure observation anyone cites, and a claim resting on one untraced report should be flagged, not quietly dropped.
Three hypotheses, none of them settled
The arc follows the toroid's field lines. This one is Uspring's. A QCW arc grows slowly enough to follow the field. A small toroid holds little charge, so its influence falls off quickly as the arc gets longer, and the arc heads for the nearest ground: floor, wall, ceiling. Outdoors there is no ceiling, so the loop goes to the ground or a wall.
The argument depends on which way the breakout points, and that can be tested. Uspring says that with a breakout facing upward "most of the field lines will go up to the ceiling and most of the arcs will follow them", and that Jan seeing the loop outdoors, with no ceiling, curling towards a house wall is "maybe not an accident". The hypothesis is more specific than "the nearest ground". The arc follows field lines whose end point is set by the room, and pointing the breakout somewhere else should move them.
Two separate pieces of evidence go against it. Jan's photographs contradict the simple version. And its main prediction, that a bigger toroid straightens the arc, has more or less been tried: davekni runs a 545 mm top load, which he notes is significantly larger than any other QCW on the forum, and reports many curved arcs and relatively few straight ones (HVF 2397). That is not a controlled experiment, but it is evidence, and it points the wrong way. The iron frame is a third: davekni was surprised that being near metal did not curve the arc towards the metal (HVF 2434).
There is an obvious objection: the arc does grow away from the topload at first. Someone asks it on the forum in those terms. If the topload's field is strong enough to steer the arc, should it not push the arc away rather than pull it back? Uspring agrees for the early arc. The field at the tip comes from the space charge just behind it and from the toroid, and at first that makes "the arc to grow straight and away from the top load". The sign of that field does not change, but its weight does: "the toroid doesn't hold much charge, so that its influence on the arc direction diminishes fast as the arc lengthens", said of a coil whose toroid is small compared with the arc it throws. In this picture the loop does not come from the topload pulling harder. The topload stops mattering, and something else takes over. That is the next hypothesis.
The arc as a distributed RC. The label came from Jan, who raised it as an objection: he had always thought of the arc as a distributed RC network, which possibly cannot produce such large phase shifts. Uspring's own picture was a chain of resistances and space charges, and he accepted the RC framing in his reply: the phase shift of a distributed RC network can be arbitrarily large, but it comes with an attenuation of voltage. He calls the whole idea speculation. The mechanism is that a phase shift builds up between the charge on the toroid and the charge at the tip, and as it approaches 180 degrees the tip is attracted back towards the toroid, which makes a loop.
That attenuation can be put in numbers. On a distributed RC line γ = √(jωRC), so the loss and the phase per unit length are equal, α = β = √(ωRC/2). Reaching 180 degrees therefore costs exactly e^-π, which leaves 4.3 per cent of the toroid's voltage at the tip, a factor of 23 down [derived]. That puts a figure on Uspring's own doubt: the loop should form just when the arc has nothing left to grow with.
Jan's own two numbers seem to contradict that last part. His coil makes "approximately 2-2.5m long sparks", and the loops he photographs are "with a diameter really less than 1 meter", both from the same thread. Using the geometry below, a metre of loop diameter is about 1.57 m of channel. So the arc reaches 180 degrees with about a metre still to go and keeps curling past it, which on a 2.5 m spark is most of a full circle. Whatever voltage is left at the tip by then, it is not zero, and "no voltage left to grow" is the part of the picture his measurements contradict. No one in the thread did this arithmetic, and it needs a better arc length than "2 to 2.5 m" before anyone treats it as established.
A spiral is a different claim, and it comes with a test. Uspring separates a meandering arc from a spiralling one, and gives the second a cause and a prediction: "Spiraling could be explained by the secondaries magnetic field. It should then always have the same helicity for every arc and it should be opposite to that of the secondary winding." You can check that by eye, with no instruments. Photograph enough arcs and either every spiral turns the same way, against the winding, or the explanation is wrong. No one in this corpus reports having checked, and it is the cheapest open question here.
The RC model also predicts how the loop size changes with frequency, and two of Jan's numbers can be checked against it. Phase per unit length goes as √f, so the length to 180 degrees goes as 1/√f. From his coil's ramp, 400 down to 350 kHz, to the VTTC's 2 MHz is a factor of √(2000/375) = 2.3, so a 1 m loop there predicts 43 cm at 2 MHz [derived, from the RC line and those two figures]. What was reported at 2 MHz is a few tens of centimetres.
That 1 m is a diameter, which is Jan's own word for it above. The scaling is a ratio, so it holds for any measure as long as it is the same one at both ends. Taking Jan's metre as a radius and comparing it with a diameter reported at 2 MHz mixes the two, and that is why it misses by a factor of two.
The model predicts a length along the channel, while what people report is a width across the room, so one line of geometry is needed to connect them. A channel that has turned through 180 degrees at even curvature is a semicircle. Its length is L = πr, so the loop's diameter is 2L/π, about 0.64 L [derived]. A loop a metre across is therefore an arc about 1.57 m long, and that is the figure to compare with a coil's reach, not the metre. No one has published both numbers for the same arc. That would be the cheapest way to test this model, and it has not been done.
No one has a mechanism for the effect of the air, and Uspring, who proposed the models, says so. Asked to explain Jan's photograph, he answers "I can't", and then: "And I also don't have the slightest idea how air temperature or humidity could cause arcs to bend or not. Does anybody have a suggestion?" Nothing tested has come back since.
Uniformity rather than absolute values. davekni has no conclusion. His best guess is that what matters is how uniform the temperature and humidity are, not their values, with the arc bending towards pockets of warmer air where the breakdown voltage is slightly lower. That last part can be tested; "uniformity matters" by itself cannot. He also notes that dielectrics matter, and that wood, which is not very conductive at low frequency, can be significantly conductive at 12 MHz, which fits CJ's wooden cabinet. Note the 12: it is thirty times a QCW's operating frequency, so it describes how the material trends, not what the cabinet is doing at 400 kHz.
What the data does not support
There is no single factor, and the temperature results are worse than inconsistent. Saying that temperature worked for CJ and did almost nothing for Jan would be too generous. Jan's result has the opposite sign:
CJ cold winter, arc bending down
lit the stove, the room warmed, it straightened heating helped
Jan heating the air to room temperature
"didn't help much" heating did not help
opening the garage door and letting outside
air in: "the looping decreased a lot" fresh air helpedBoth of them heated the room, and it worked for only one.
Jan's second line is not really a temperature result. His words are "Even though it was relatively cold (~10 degrees C), the looping decreased a lot", so he treats the cold as something the improvement happened in spite of. What he changed was the air itself, swapped with outdoor air, and he says so. No one here claims that colder is straighter. One builder says heating helped, and the other says heating did not help but fresh air did.
The one pressure observation has no traced source, and no one has repeated it.
The two results can be made to agree, but only through a quantity neither of them measured. Heating a room without adding water lowers its relative humidity, so CJ's stove made the air drier. Outside air at ten degrees coming into a warmer garage warms up, and its relative humidity drops as it does, so Jan's open door also made the air drier. Plotted against relative humidity instead of temperature, both results point the same way, and it is the direction Jan reports himself: drier air, straighter arcs.
That is a hypothesis, not a finding, and it rests on two weak supports: a correlation that its own author calls weak and says he does not really understand, and an assumption about the absolute humidity of the outdoor air at Jan's garage that no one measured. Its merit is that it is the only explanation that fits both results without throwing one away, and it can be tested with the cheapest instrument mentioned anywhere here.
Almost none of this is a direct experiment. Nearly all of it is "I changed the surroundings and it behaved differently", with several things changing at once. None of these factors has been isolated, so everything in the next section is a rule of thumb, not a conclusion.
What to do with it anyway
- If your arc bent after a move or a cold snap, look at the room before you touch the tuning. This is the most useful advice here.
- Do not compare your results to somebody's video without knowing the conditions it was shot in.
- If it loops down towards the base, a bigger toroid is hypothesis one. Try it knowing that the largest top load on the forum produces mostly curved arcs, so what evidence there is goes against it.
- Check whether it is actually branching rather than wandering. Branching has its own causes: the lower pole, too fast a ramp, a corner in the ramp, or residual voltage left on the bus.
What would close it
- One factor isolated. One coil, one room, change only the humidity, or only the pressure, and film it on one camera.
- The frequency dependence on one machine, instead of comparing different machines. The square root above says what to expect.
- Move the breakout point instead of the top load. Uspring suggested this to davekni, and it is better controlled than swapping toroids because it changes the field geometry without touching the tuning: a breakout rod a metre long, or one set off centre, or one at an angle. His reason it is safe to try is that "Adding extra capacitance to the top likely won't affect tuning very much, since the coupling is so large". It also tests the field-line hypothesis directly, since that hypothesis says the direction the breakout faces decides which surface the field lines end on. Not yet tried.
- A hygrometer, the cheapest item on this list. Every room result above is a temperature story from people who were not measuring humidity. The two temperature results point in opposite directions, while their humidity readings, had anyone taken them, would point the same way. Log relative humidity next to each run and the leading hypothesis gets its first real test.
- Numbers. Every observation above is qualitative, and that is still the gap. The two figures here, the 4.3 per cent and the 43 cm, come from a model, not from measurements, and the second is the model's only successful prediction so far.
One good set of any of these measurements would answer a question the community has not yet answered.
The QCW department diagram draws its arc straight, which is the good case, not the usual one. The records page collects the builds that got there.